论文部分内容阅读
在使用支持向量机(SVM)分类时,存在以下两个问题:一是当存在噪点时,分类的精度低;二是对大规模样本集,训练时所需内存空间较大,运行时间较长.针对以上问题,给出一种基于具有距离性能的核函数的减样方法,称为删减法(DRM).该方法定位定量分析了噪点及多余样本点的一般比例.在应用时,分三步进行:首先根据小概率原理给出一小阈值删除噪点;然后给出一个较大阈值减去同类中心附近的大量多余的样本点;最后以另一个大的比例减去位于距异类中心较远的对分类不起作用的样本点,以便提取具有代表性的边界向量.试验结果检验了该方法的有效性,即,既减少了训练时间,又提高了分类精度.
There are two problems in using SVM: First, the classification accuracy is low when noise is present; second, for large-scale sample sets, the memory space required for training is large and the running time is longer Aiming at the above problems, this paper presents a subtraction method based on the kernel function with distance performance called subtraction method (DRM), which locates and quantifies the general proportion of noise and redundant sample points. Three steps: First, a small threshold is used to remove noise based on the principle of small probability; then a larger threshold is subtracted from the large number of extra samples near the center of homology; and finally, Far away from the classification of sample points in order to extract a representative of the boundary vector.The experimental results verify the effectiveness of the method, that is, both to reduce training time, but also improve the classification accuracy.